
Company interview guide
Belvedere Trading Machine Learning Interview Guide (2026)
For ML, AI and research roles: a timed at-home test, a virtual probability and resume round, then a half-day final on signals, statistics, options intuition and fit.
Belvedere Trading's machine learning interview loop has 8 reported stages, from application & recruiter screen to final round (super day): depth, coding & fit; typical length: About 3-8 weeks.
Roles covered: Research Scientist, Data Scientist, ML Engineer and Applied Scientist. The question bank holds 3 questions reported from Belvedere Trading interviews, 2 of them free to solve in your browser.
- Interview process length
- About 3-8 weeks
- Stages
- 8
- Reports linked to bank questions
- 3
- Bank report dates
- Oct 2025–Oct 2025
- Questions in bank
- 3 (2 free)
- Last reviewed
- October 2026
How the Belvedere Trading interview loop works
Apply to one position (the firm says position, title and location cannot change after an offer), a resume review that does not screen on school, major or GPA, then a role-specific timed at-home test: a probability and statistics test for research and trading candidates, a coding challenge for developers. Some research and engineering candidates then get a take-home project. A virtual first round with a researcher, trader or developer covers your resume and technical questions, and a half-day final round (in Chicago with a reception, or on Zoom) has several interviews with quants and a trader, ranging over signals, statistics, options and volatility, coding and fit. The firm says decisions usually follow within a few weeks of the final round; campus first rounds start in late August and finals run in September. Mental-math drills and market-making games reported at the final are trader-track only. The outline rests on few research-role reports, so confirm each round with your recruiter.
The Belvedere Trading interview process, stage by stage
01Application & recruiter screen
Online application with your latest resume, reviewed on a rolling basis; often a short recruiter or HR call on background and interest
You apply to exactly one position, and the firm states that the position, title, location and internship dates cannot change once an offer is made. Resume review does not filter on school, major or GPA. The entry-level quant researcher posting asks for a master's or PhD in a quantitative field, about two years of research experience, strong Python or C++, and work with large datasets and statistical modelling; it does not mention machine learning by name. Campus roles open in early August for the next summer.
What it tests
- Why this firm and why research rather than trading or engineering
- Evidence of statistical modelling on large, noisy datasets
- Python or C++ fluency and collaborative coding habits
- Logistics: track, start date, sponsorship (offered for research and software roles)
How to prepare
- Choose the track deliberately: the test you sit next depends on it and you cannot switch later.
- Phrase ML work in the posting's terms: hypothesis, backtest, validation, monitoring for drift.
- Apply early in the August window; review is rolling and may have an early deadline.
The firm's careers and campus pages set the one-position rule, rolling review and timeline; 3 trader-track reports describe a recruiter or HR call, a step shared by every track.
02Timed test: probability, statistics & regression (research track)
At-home, timed; compiled accounts describe a multiple-choice concept test drawn from a pool of about 22 items, mostly spot-the-false-statement; trading-track sittings are reported at about 90 min with about 15 probability questions plus a short code-a-statistic part
The firm describes a timed probability and logic test for quantitative candidates. Compiled accounts of the research version list multiple-choice items on ridge versus lasso, what happens to standard errors and R-squared when well-fit points are deleted, the position of the first ace in a shuffled deck, a minimum-variance hedge across three correlated assets, option sensitivities and variance additivity for volatility. A related at-home format pairs quick probability with coding a statistic such as a mean, standard deviation or Sharpe ratio, where the sample versus population denominator is the trap.
What it tests
- Regularization facts: ridge shrinks but never zeroes, lasso selects
- Regression diagnostics: residuals, standard errors and R-squared under data changes
- Card, coin and dice probability with symmetry and conditioning
- Variance of linear combinations and hedge ratios from a covariance matrix
- Turning a statistic into exact, edge-case-safe code
How to prepare
- For each 'which statement is false' item, look for the one counterexample instead of proving every option.
- Know why deleting small-residual points raises the standard error and lowers R-squared.
- Derive the minimum-variance hedge as a regression coefficient; it is the same algebra.
- When coding a statistic, decide n versus n-1 up front and guard empty or constant series.
A test appears in 2 research-role reports (2020, 2026) and 4 trader-track reports with math or probability content, a shared step; item content comes from compiled accounts of the research test. The firm's pages describe the timed test.
03Timed coding challenge (developer track)
At-home, timed; compiled accounts describe 3 problems over about 30 min + 1 h + 1 h; 2025-2026 reports also describe a recorded coding exam and a two-question medium-to-hard set
Engineering-leaning ML candidates usually sit the developer coding challenge, which the firm says tests fundamental tech concepts. Compiled accounts name a recurring set built on trading mechanics: an order-book price ladder with add, delete and level-clear queries, a running trade aggregator, arbitrage across nested stock bundles, and maximum profit with at most K trades. Sample tests are sparse, so misreading the rules costs more than algorithmic depth. One 2025 candidate got a system design question in the same set.
What it tests
- Order-book bookkeeping: best bid and ask, depth, cancels
- Aggregating a trade stream with tie rules
- Bounded-transaction dynamic programming on a price series
- Graph or recursion over nested baskets
- Clean, efficient code in C++ or Python
How to prepare
- Read every tie-breaking and ordering rule twice before coding; few sample cases are given.
- Write your own edge-case tests: empty book, cancel of unknown id, equal prices.
- Be ready to explain an alternative algorithm; one 2026 interviewer reviewed the candidate's recorded test code.
Three software engineer reports (2025-2026) describe a recorded coding exam, a take-home or a two-question set; problem content comes from compiled accounts. The firm's pages name a timed coding challenge for developers.
04Take-home project (some teams)
A technical take-home project after the test, then reviewed or extended live; reported for a 2026 quant researcher and a 2025 engineer
Not every candidate gets one. A 2026 quant researcher did an online assessment and then a technical take-home before four interviews; a 2025 engineer later spent part of a three-hour interview revising and extending the take-home code. Trader-track candidates describe a coding project plus a discussion of it inside the final round, so the project can also appear there.
What it tests
- Turning an open-ended data or coding task into a clean, reproducible result
- Explaining design choices and what you would test next
- Extending your own code live when the interviewer changes the spec
How to prepare
- Write it as if a colleague will extend it tomorrow: small functions, a short readme, tests.
- For a research task, state the validation scheme and the result you would not trust.
- Expect follow-up questions on any shortcut you took; keep notes on why.
One 2026 quant researcher and one 2025 engineer report a take-home; compiled accounts of the research loop do not list it, so treat it as team-dependent.
05Virtual first round: resume & probability
Video call with a researcher, trader or developer; compiled accounts put it at 45-60 min; technical and behavioral questions
The firm describes a virtual first round with a current researcher, trader or developer covering technical and behavioral questions. A 2025 quant researcher's first round was a resume discussion plus a probability question; another research candidate called the expected-value question hard. Compiled accounts list dice, coins and cards, races and waiting times such as first to roll a six or pattern races, Bayesian updates such as the double-headed coin, and fast expected value.
What it tests
- Expected value by conditioning and linearity
- Bayesian updates after an observation
- Pattern races and waiting times in coin and dice sequences
- A clear account of your own research or project
How to prepare
- Give a quick estimate first, then make it exact; say the setup out loud.
- Use first-step analysis for races and waiting times, indicators for counts.
- Prepare a five-minute walk through one project with the numbers you would defend.
Free reported questions for this stage
- Josephus Problem: Last Person Standing in a CircleProbabilityMediumSolve
- Expected Balls Left in the Banach Match ProblemProbabilityHardSolve
Two quant researcher reports (2025) describe a resume and probability or hard expected-value round; topic list from compiled accounts. The firm's campus page describes this virtual round.
06Research round: signals, statistics & validation
One or more interviews with quant researchers, in a second technical round or the final; open-ended, exploratory discussion
A 2025 quant researcher's second round asked them to build a trading strategy and analyse possible signals; a 2026 candidate described the senior-quant interviews as exploratory, testing intuition on fundamental quant questions and their own experience. The posting stresses backtesting, validation, monitoring and adjusting models for data drift. Compiled accounts add order-book features such as the micro-price from bid and ask imbalance and minimum-variance hedging, which fit here as signal and risk questions.
What it tests
- Designing a signal: hypothesis, features, horizon and how you would measure it
- Backtest pitfalls: look-ahead, leakage, overfitting and multiple testing
- Time-ordered validation (walk-forward, purging) on non-stationary data
- Regularization and shrinkage for weak, noisy signals
- Spotting signal decay and drift after deployment
How to prepare
- Start any strategy question with the hypothesis and the cost of being wrong, then the data.
- Name how many variants you tried and how you corrected for it before quoting a Sharpe ratio.
- Have one example of a model that degraded in production and how you caught it.
- Practise deriving the micro-price from order-book imbalance and explaining what it predicts.
Two quant researcher reports (2025-2026) describe a strategy and signal-analysis round or exploratory senior-quant interviews; the posting supplies the validation and drift focus.
07Options & volatility intuition
Questions inside first-round or final interviews; reported for quant researchers, not a separate sitting
Research candidates report options questions even though the firm assumes no finance background for campus hires. One 2026 candidate was asked why equity volatility surfaces skew toward puts while commodity surfaces skew toward calls; a 2025 candidate got Black-Scholes pricing, skew and smile, and term structure. Compiled accounts add option sensitivities, butterfly and convexity arbitrage across strikes, and variance additivity under geometric Brownian motion. The firm's experienced-hire page says researchers focus on option modelling across indices, commodities and equities.
What it tests
- Signs of the main option sensitivities
- Why volatility skew differs between equity and commodity options
- No-arbitrage: call prices convex in strike, butterfly spreads
- Volatility scaling with time and the term structure
How to prepare
- Learn the economic story behind skew (crash risk versus supply shocks), not only the shape.
- Be able to sketch the Black-Scholes assumptions and say which ones fail in practice.
- Skip market-making quoting drills unless you are also on the trader track.
Two quant researcher reports (2025, 2026) describe volatility skew, Black-Scholes and term-structure questions; extra topics come from compiled accounts of the research loop.
08Final round (Super Day): depth, coding & fit
Half day, in Chicago with a reception or on Zoom; a 2026 quant researcher had four 30-60 min interviews with three senior quants and a trader; engineers report up to a 3-hour technical block
The firm describes a half-day final with hands-on practice and in-depth discussion of your skills, meeting several team members. The research final mixes exploratory quant questions, options intuition and your experience; trader and engineering finals add a coding problem or coding project and a behavioral interview. A 2025 engineer's three-hour block covered live coding, system design and extending the take-home. The firm values teamwork, coachability and an entrepreneurial spirit.
What it tests
- Hardest variants of the probability and statistics material
- Depth on your research or project under follow-ups
- Coding with attention to correctness and performance
- Teamwork, coachability and interest in markets
How to prepare
- Plan for back-to-back interviews; keep one fresh example per value the firm names.
- When a senior quant pushes back, update your estimate openly rather than defending it.
- Ask about the team's models and data; curiosity is part of what is assessed.
- If you hear nothing after a few weeks, a polite follow-up is fine.
A final half day appears in 1 quant researcher report (2026) and 7 trader-track reports (2021-2026), a shared step; content differs by track. The firm's pages describe the half-day final and reception.
Questions and themes reported most often
- Hard expected-value and Bayes questions in the first round (2 reports, virtual first round: resume & probability)
- Volatility skew, Black-Scholes and term structure (2 reports, options & volatility intuition)
- Build a trading strategy and analyse candidate signals (1 reports, research round: signals, statistics & validation)
- Spot the false statement about ridge, lasso and regression diagnostics (timed test: probability, statistics & regression (research track))
- Dropping well-fit observations: what happens to standard errors and R-squared (timed test: probability, statistics & regression (research track))
- Minimum-variance hedge across correlated assets (timed test: probability, statistics & regression (research track))
- Code a mean, standard deviation or Sharpe ratio exactly (timed test: probability, statistics & regression (research track))
- Order-book price ladder and running trade aggregates (timed coding challenge (developer track))
What Belvedere Trading asks about
- Probability & expected value25%
- Statistics & regression22%
- Coding (Python or C++)20%
- Signal research & model validation15%
- Options & volatility intuition10%
- Behavioral & fit8%
Belvedere Trading interview prep checklist
- Can say which ridge, lasso and OLS statements are false and why, without notes
- Can derive a minimum-variance hedge and the variance of aX + bY from a covariance matrix
- Coded mean, sample standard deviation and Sharpe with empty and constant inputs handled
- Solved a set of EV, Bayes and coin-pattern problems out loud under a timer
- Can explain walk-forward validation, look-ahead bias and multiple-testing corrections for a signal
- Can explain why equity skew favours puts and why call prices are convex in strike
- Prepared one project deep dive and one example of a model that degraded after deployment
- Confirmed with the recruiter which test, take-home and final-round format apply to you
Tips
- Confirm your track early: research candidates sit a probability and statistics test, developers a coding challenge.
- Treat regression and regularization as exact facts; the research test asks which statement is false.
- Drill expected value, Bayes updates and pattern races until the setup is automatic.
- Know covariance algebra well enough to derive a hedge ratio or the variance of a portfolio in a minute.
- Prepare one research project told end to end: hypothesis, data, time-ordered validation, what failed.
- Learn options intuition (sensitivities, skew, convexity) even for an ML role; researchers here report it.
- For engineering-leaning roles, practise order-book and stream-aggregation problems with careful tie rules.
- Skip mental-math speed drills and market-making games unless you also apply to a trader role.
See the rest of the Belvedere Trading loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Belvedere Trading loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Belvedere Trading interview FAQ
How many interview rounds does Belvedere Trading have for ML roles?
Candidates report 8 stages: Application & recruiter screen, Timed test: probability, statistics & regression (research track), Timed coding challenge (developer track), Take-home project (some teams), Virtual first round: resume & probability, Research round: signals, statistics & validation, Options & volatility intuition and Final round (Super Day): depth, coding & fit. Apply to one position (the firm says position, title and location cannot change after an offer), a resume review that does not screen on school, major or GPA, then a role-specific timed at-home test: a probability and statistics test for research and trading candidates, a coding challenge for developers. Some research and engineering candidates then get a take-home project. A virtual first round with a researcher, trader or developer covers your resume and technical questions, and a half-day final round (in Chicago with a reception, or on Zoom) has several interviews with quants and a trader, ranging over signals, statistics, options and volatility, coding and fit. The firm says decisions usually follow within a few weeks of the final round; campus first rounds start in late August and finals run in September. Mental-math drills and market-making games reported at the final are trader-track only. The outline rests on few research-role reports, so confirm each round with your recruiter.
How long does the Belvedere Trading ML interview process take?
Reported timelines run About 3-8 weeks.
What does Belvedere Trading ask in ML interviews?
By reported question mix: Probability & expected value 25%; Statistics & regression 22%; Coding (Python or C++) 20%; Signal research & model validation 15%; Options & volatility intuition 10%; Behavioral & fit 8%. The most repeated themes: Hard expected-value and Bayes questions in the first round; Volatility skew, Black-Scholes and term structure; Build a trading strategy and analyse candidate signals.
How should I prepare for the Belvedere Trading ML interview?
Confirm your track early: research candidates sit a probability and statistics test, developers a coding challenge. Treat regression and regularization as exact facts; the research test asks which statement is false. Drill expected value, Bayes updates and pattern races until the setup is automatic.
Are there free Belvedere Trading interview questions?
Yes. LeetML has 3 questions reported from Belvedere Trading interviews, and 2 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- Akuna CapitalFor ML, AI and research roles: a proctored Python coding test, a camera-on math test, probability interviews, then a final round on statistics, ML and your research.
- OptiverFor ML, AI and research roles: a long timed assessment battery, a live betting game with a trader, a probability whiteboard, a P&L-graded take-home and a blunt fit final.
- Susquehanna (SIG)For ML, AI and research roles: a timed probability test, regression-heavy technical calls, a trade-data take-home and a long onsite.
- IMC TradingFor ML, AI and research roles at a global options market maker: a track-split timed test, probability and ML-fundamentals rounds, and a fast superday.
- Old Mission CapitalFor ML, AI and research roles: a 25-minute, 35-question probability speed test, a behavioral round, then probability, statistics and coding rounds at a multi-asset market maker.
- DRWFor ML, AI and research roles: a timed Python and ML test, then probability, regression and an ML design or research deep dive over a two-day final.